Extracting Supply Chain Information from News Articles Using Large Language Models: A Fully Automatic Approach

Installation

This codebase is only tested on Windows 11.

1. Download uv
2. Open cmd from directory
3. Type uv sync
4. Run each jupyter notebook, using the created virtualenv

Requirements

Relation Classification Dataset

To create 'ManualDataset' and 'ManualReducedDataset', the dataset of Wichmann et al. is needed. Go to https://github.com/pwichmann/supply_chain_mining and follow the instructions on obtaining the dataset.

Huggingface Token

To use LLama-3-8B-Instruct, you need an access token from huggingface and the permission to use the model from Meta. Here are the necessary steps:

1. Go to https://huggingface.co, and create an account.
2. Then, go to https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct, and acquire access to the model. This may take several days.
3. If access to the model is granted, go to https://huggingface.co/settings/tokens and create a new access token. Select 'Read' from 'Token type'.
4. The access token will be shown only once after it is created (starting with 'hf_'). Copy and paste it in the .env file as 'ACCESS_TOKEN', such as ACCESS_TOKEN=hf_...

Case Study Dataset

To obtain the case study dataset (mining.json, mining_processed.json), please contact the author ([email protected]).